Written before the test started
Forward-paper protocol
One strategy passed. This is what it now has to survive. The whole point is that you are reading it before the result exists.
I committed this plan in July 2026, before the first day of data. How the strategy is built is not published, because it is still being tested. What is published is every condition I will judge it by. That is the half a person can quietly bend afterwards, so that is the half worth fixing in advance.
Level 1: does the test match the study
When decisions are made, when orders fill, how costs are counted, and keeping it separate from everything else: all of this has to match the backtest exactly. There is no allowance here. Any difference is a bug to fix, not a gap to explain. If the paper trading and the backtest can disagree, then the paper trading is measuring something else.
Orders are decided one evening and filled the next, at the following day's opening price, which is what the backtest assumed. Being one day out here is one of the commonest and hardest-to-spot ways paper trading drifts away from the study it is supposed to be checking.
Level 2: when to stop
Hard kill
If the paper account falls past the loss limits I wrote down, the strategy is stopped for good. Not paused, not retuned, not reviewed. It is finished, and anything replacing it starts as a fresh idea with a fresh plan.
Soft flag
If it does worse than the worst the backtest actually showed, I have to write up a review. This does not stop anything and it does not let me change the strategy. It just forces me to record what happened.
Level 3: the two result dates
Both dates were fixed in advance.
Month six · ~February 2027
Needs a minimum number of clean monthly rebalances, no stop triggered, and results inside the range the backtest showed. Passing only opens the question of using real money and nothing else. It does not prove the strategy works. Six months cannot prove that, and the plan says so in writing so a later version of me cannot pretend otherwise.
Month twelve · ~August 2027
The first full year on data the strategy has never seen, compared against what the backtest showed. Honestly, that is still one year.
What I will not count as evidence
I have ruled out, in advance, any comparison against a benchmark over less than twelve months. Over a few months that comparison mostly reflects which way the market happened to go. I would be tempted to quote it if it looked good and ignore it if it did not, so it is easier to throw it out now than to resist it later.